Helping creators stay in control when using generative AI
4 November 2026
Picture a designer uploading an unpublished illustration to an AI tool to explore different backgrounds. They might welcome suggestions, but not want their original work reused to train the system. A filmmaker, meanwhile, might be willing to share material, provided their contribution remains identifiable and they receive credit or payment where appropriate.
These different needs are central to SafeGenAI. The two-year project, coordinated by the UvA and supported through TKI CLICKNL’s public–private partnership programme, brings together the university, Privacy First, the Netherlands Institute for Sound & Vision and STORM Postproduction. It combines computer science with research into creativity, entrepreneurship and the legal and practical concerns of creative professionals.
Keeping creative contributions visible
At the UvA’s Informatics Institute (IvI), Ana Oprescu and Sander Klous from the Complex CyberInfrastructure (CCI) research group will develop technical safeguards. Project coordinator Oprescu will investigate cryptographic techniques that could let a designer receive AI assistance without revealing an unpublished illustration to the system. She will also explore how to track the origins of creative material and how it is used to train AI models or generate new content. Klous will lead research into auditing: recording source material, prompts and human decisions to make AI-assisted work easier to inspect.
Tom van Engers will contribute AI and law expertise to the policy-related research and connect the project with the ELSA Labs network, which brings together ethical, legal and societal perspectives on AI.
Is AI doing more than imitating?
Protecting creative contributions also requires a better understanding of what AI produces. Monika Kackovic, from Entrepreneurship & Innovation at the UvA’s Faculty of Economics and Business (FEB), will coordinate the conceptual design of new measures of AI creativity.
The team, including Thanos Eftymiou and Stevan Rudinac from FEB, will look beyond whether an AI-generated image resembles a particular artistic style. Can a model produce something original and meaningful? Can it communicate an abstract idea, such as love or migration, rather than reproduce familiar visual patterns?
Using art-historical data and possibly material from Sound & Vision’s collections, the researchers will develop ways to assess these questions. Expert evaluations and audience responses will help test whether the measures capture qualities that people recognise as creative.
Developed with creative professionals
Designers, audiovisual makers and cultural institutions will help shape and evaluate the research through interviews, workshops and practical testing. A prototype interface will let creators explore different choices, from keeping material private to permitting reuse under specified conditions.
For the designer experimenting with an unpublished illustration, the aim is to make AI assistance less of a trade-off: exploring new ideas while retaining control over the original work. SafeGenAI will investigate what technical tools and working practices are needed to make that possible.